Adaptive parameter compensation for robust hands-free speech recognition using a dual beamforming microphone array

This paper proposes a parameter compensation technique to improve the performance of hands-free speech recognition in noisy environments. The technique employs a microphone array to perform spatial filtering on the noisy input signals. A dual beamformer is used to form an estimate of both the desired speech signal and also the combined noise signals. Additional enhancement to the desired signal features is performed using the estimated noise feature vectors. The parameter compensation is seen to provide significant performance improvement in a simple speech recognition application